基于GANs-LightGBM的序贯三支异常用户检测研究

针对网络中异常数据类别分布的不平衡性和异常用户检测代价的敏感性,在序贯三支决策框架下,提出了一种基于生成式对抗网络和集成学习模型的异常用户检测方法.利用生成式对抗网络(generative adversarial nets,GANs)模型对异常/非异常数据进行类别平衡,并在多层次多粒度的特征空间下训练LightGBM模型,持续地处理不确定域的样本以识别异常用户.实验结果表明,与传统的机器学习算法相比,该方法在异常用户检测中具有较高的AUC值和较低的检测代价.

Research on sequential three-way abnormal user detection based on GANs-LightGBM

Aiming at the class imbalance of abnormal data distribution in the network and the cost-sensitive of abnormal user detection, this paper proposes an abnormal user detection method based on a generative adversarial net and an ensemble learning model under the framework of sequential three-way decision. The GANs model is used to balance the class of abnormal/non-abnormal data, and the LightGBM model is trained in a multi-level and multi-granular feature space to continuously process samples in uncertain region to identify abnormal users. Experimental results show that compared with traditional machine learning algorithms, this method has higher AUC value and lower detection cost in abnormal user detection.